Large Vocabulary Continuous Speech Recognition

Sensory is a pioneer in the use of embedded neural network-based speech recognition and has become the industry leader in optimizing and engineering speech recognition software with small footprints and minimal MIPS. This extensive experience and continuous innovation have led to the first embedded large vocabulary continuous-speech recognizer (LVCSR) with state of-the-art cloud performance. Unlike voice recognition software often used with smartphones and mobile devices, such as with a voice assistant mobile app, as well as with IoT (internet of things) enabled technologies (Alexa, Google Assistant, Siri, Cortana), Sensory’s solution is embedded and doesn’t require a wifi connection. 

Image Alt Text


Many applications don’t need or want to rely on cloud-based connection to do high-performance speech recognition. Others seek a client/cloud distributed system with optimal performance. The market concerns regarding privacy, performance and bandwidth are driving more processing to the edge. TrulyNatural (TNL) is the natural language speech recognition solution for these needs. Sensory’s TrulyNatural engine is ideal for a multitude of purposes, and it boasts true security since voice data never leaves a device, nor is it stored in the cloud. This means all personal information is kept safe thanks to full user data privacy. TNL can be integrated into devices of all kinds and will continue to help appliance and voice assistant manufacturers create powerful, accurate, and secure devices. With voice biometrics and other security options, homeowners can even control who can use the devices.

Image Alt Text

Key Features

State-of-the-art, noise-robust, and an order of magnitude smaller than comparable solutions, now includes speech-to-text!

Fast- Highly optimized neural network reduces MIPs. Running on device eliminates cloud communication delay.
Accurate- Best-in-class for embedded applications. Outperforms published benchmarks for cloud-based solutions.
Customizable- The recognition vocabulary can be tailored to a specific task or domain.
Private & Secure- Running on-device avoids the need to transfer sensitive information to the cloud
Natural- Allows users to speak naturally to their devices. Avoids need to learn restricted command sets


Sensory’s TrulyNatural software developer kit (TNL SDK) provides a state-of-the-art, embedded, small-footprint, and speaker-independent, large-vocabulary continuous-speech recognizer that is both scalable and customizable for consumer electronics products and is easy to integrate into applications. It is highly scalable in size and features and can run on a variety of operating systems or be ported to custom platforms. It’s also available in the cloud with the cloud solutions marketplace and transforms as your needs do for practically limitless scalability.

Request the TrulyNatural™ product brief to learn more.

Image Alt Text)

A Multitude of Use Cases

TrulyNatural™, Sensory’s Natural language processing (NLP)  and natural language understanding (NLU)  platform, isn’t limited to smart home virtual assistants. The powerful NLP & NLU AI software may be used to identify a person’s voiceprint for security purposes, and when combined with facial recognition technology, this can make for a practically impenetrable security system for banking, enterprise, and  mobile applications.

Conversational chatbots enabled with natural language voice recognition can assist customers with simple queries, and interactive voice response (IVR) systems enhanced with an embedded NLU that is domain specific can help customers reach the appropriate support agent for their issues faster, improving the customer experience.

Sensory has enabled QSR food kiosks with natural speech recognition, eliminating the need for a touch-screen interface, an increasingly valuable feature due to the Covid-19 pandemic. Sensory’s technologists work with clients globally to create natural interactions with customers–accounting for regional differences in accent, pronunciation and dialects.

Image Alt Text)


Embedded NLU and speech recognition algorithms are important for conversational AI devices to be able to understand, not only human vocabulary, but also context behind it for advanced functions. Sensory’s on-device assistant is able to outperform the Amazon cloud assistant in a domain-specific environment, where it’s able to easily reject irrelevant sounds picked up by the microphone to reduce error rates. Results show that Sensory’s embedded LVCSR & NLU approaches the cloud-based Google Assistant’s 97% English word accuracy with an accuracy of rate of 96%. The ability of domain-specific speech recognition technology to parse accents, recognize mispronunciations, determine voice input from background noise, and perform rapid computational linguistics makes it an ideal solution for a variety of applications from food ordering to voice controlled home appliances.

Request the Vocalize Voice Assistant Accuracy Report

Image Alt Text)

Stay Connected

Get the latest speech recognition and artificial intelligence resources and industry updates.